Disentangling age, period, and cohort effects
When interested in changes of psychological variables across the life-span, researchers are confronted with complex methodological issues associated with the simultaneous estimation of age, period, and cohort effects (APC). These three components of change were defined as follows (Study III): Age effects are effects of growing older (for instance, climbing stairs is easier for a young than for a very old person; role changes like those associated with becoming parents, etc.). Period effects are societal, historical, or cultural changes simultaneously influencing all cohorts. Cohort effects are “the replacement of
(happier or less happier) cohorts born early in the 20th century by cohorts born later in the same century” (p. 298).
By aiming to examine cultural differences in change trajectories, we used an alternative to longitudinal studies, which is an intra-cohort aging approach to change, also called a cross-sectional sequential design (Baltes, 1968; Glenn, 2005). “Intra-cohort aging summarizes the net results of individual-level change and is, therefore, a conservative aggregate measure of what is happening at the individual level” (Danigelis, Hardy, & Cutler, 2007, p. 813). This approach is based on the basic claim that members of a birth cohort share special characteristic features due to the unique socio-historical experiences they had in their formative years (i.e., during socialization) (e.g., Ryder, 1965). The design makes it possible, by building on an age by cohort table, to follow birth cohorts as they grow older in order to disentangle effects of aging from period and cohort effects. The interaction effects of age or cohort and time of measurement (i.e., period), however, must be examined in separate analyses due to an identification problem (Mason, Oppenheim, Mason, & Winsborough, 1973).
Methodologically, the identification problem is a multicollinarity situation due to the fact that if one has the age of a respondent and the period when the survey took place, the third variable, namely, year of birth, can be perfectly calculated. For this statistical reason, it was believed until recently that confounded APC effects could not be estimated within a single statistical model (Glenn, 2005) without imposing one or more constraints (e.g., by constraining the coefficients of two periods to be equal). The results, however, will always depend on the constraints used and there is no empirical test that can give certainty whether the right ones were chosen (Tu, Davey Smith, & Gilthorpe, 2011). Thus, although several empirical approaches have been proposed in order to overcome or at least avoid the identification problem (e.g., Mishler &
Rose, 2007; Tu et al., 2011; Yang, Fu, & Land, 2004; Yang & Land, 2006), all these new APC approaches “have primarily been employed by their authors and have yet to be widely adopted or evaluated by other researchers or methodologists” (Harding, 2009, p. 1451) or “cannot or should not be used to recover the underlying age, period, and cohort effects” (Luo, in press, p. 1). For these reasons, a graphical cross-sectional sequential design was used in Study III, which looks at interactions between two of the three APC components at a time.
4.1. Cultural differences in the relationship between age and subjective well-being
People’s well-being across life span has caught the interest not only of psychologists but also of demographers and economists (Easterlin, 2006).
Related studies examine special populations like the elderly (e.g., Gana, Bailly, Saada, Joulain, & Alaphilippe, 2012), investigate changes in satisfaction over the life circle of parenthood (Powdthavee, 2009), look at gender differences
(Inglehart, 2002), or focus on the depriving effect of youth unemployment on their life satisfaction (Realo & Dobewall, 2011). Most importantly, there is an ongoing debate of whether SWB is U-shaped in life (Blanchflower & Oswald, 2008; 2009; Glenn, 2009; Sutin, Terracciano, Milaneschi, An, Ferrucci, &
Zonderman, 2013; Yang, 2008) – that people in midlife would be generally less happy than those at younger and older ages – and about the degree to which this trend can be generalized cross-culturally (Baird, Lucas, & Donnellan, 2010;
Deaton, 2008). If SWB does indeed have a universal age trajectory, it would have wide-ranging implications for policymakers and scientists in a world of aging societies (United Nations, 2010), whilst a rejection of this claim would be even more important to take into account when making culture-sensitive predictions of societal development on these measures.
Study III was carried out in order to compare the relationship between age and life satisfaction in four European nations: Estonia, Finland, Latvia, and Sweden. The age representative sequential (repeated) cross-sectional dataset was produced by merging EVS, WVS, and ESS. This way it was possible to investigate changes over a period of up to 27 years. The study makes an important contribution to the literature because studies at one specific point in time alone cannot explain why age-life satisfaction trajectories differ across countries (cf. Deaton, 2008).
The mean levels of life satisfaction in the two Nordic nations showed an almost flat trend, whilst in the Baltic countries they varied considerably. Like earlier research (Diener & Oishi, 2004; Inglehart et al., 2008; Veenhoven, 1996;
White, 2007), we observed consistent mean level differences between Estonia-Latvia and Finland-Sweden. More specifically, in the Nordic nations there was a flat trend, at a comparatively high level. In the Baltic countries the mean levels of life satisfaction varied considerably – decreasing from 1990 to 1996, then steadily increasing until 2006/2007 and finally again slightly decreasing (2008/2009) – driven by immense political and socioeconomic changes in these nations during the last two decades (Inglehart et al., 2008). Even today, people in Estonia and Latvia report, on average, two scale points lower life satisfaction than their neighbors in Finland and Sweden.
In the two Nordic countries, the relationship between age and life satisfaction was virtually zero. Unlike in Finland and Sweden, the relationship between age and subjective well-being in Estonia and Latvia was best described as curvilinear, with younger and older people having higher levels of life satisfaction. Study III also found – in line with many other studies (see George, 2010, for a review) – little evidence for declining life satisfaction in old age.
If the observed relationship between age and life satisfaction did indeed appear primarily due to age effects, there should be a universal age-related intra-cohort change, whereas the between-cohort mean level differences in SWB should be minor. Our findings, however, showed that the observed changes in life satisfaction across the life-span in Estonia and Latvia were not due to age-related changes per se but rather to an interaction of cohort (e.g., the
29
later born cohorts having high social optimism and no Soviet-time memories) and period effects (e.g., the improved economic situation).
Putting the results into the big picture, Study III serves as an example for the same historical event (such as the restoration of independence in Estonia and Latvia) resulting in different outcomes for different cohorts experiencing the same event but at different life stages (the “losers” vs. the “winners” of transition). This is alarming, because the current economic crisis, and, in particular, recently high youth unemployment in Europe, has the potential to produce a “lost generation” with permanently lower mean SWB levels than earlier born cohorts (Realo & Dobewall, 2011; Sutin et al., 2013). In such a situation, within-nation trajectories of well-being across age may change if an historical event exerts a lasting effect on (young) people’s well-being.